Uploaded October 2026 | Updated November 2026, 21 hours ago
📝 I had a blast catching up with my friend Gabriel Bernadett-Shapiro, a Distinguished AI Research Scientist at SentinelOne, after we ran into each other at Unprompted. We geeked out over how reverse engineering went from the most expensive skill in security to totally cheap in about a year, including how Gabe got root on his own TV in a couple of hours and found a hard-coded detector for a baby crying. We also ended up genuinely disagreeing about who should get access to the most capable cyber models, which was a lot of fun.
🧭 What we talk about:
🔓 "Reverse engineering used to be the most expensive skill in security": How AI agents made it cheap almost overnight, like the researcher who used Claude Code agents to turn Call of Duty: Modern Warfare II into a playable build in about three months. Plus what happens to software businesses when you can't safely ship a binary, and Gabe's prediction that we'll end up spending tokens instead of license fees.
📺 "I got root in my TV": Gabe had a model statically analyze his TV's firmware update, found a way in within about two hours, and turned up a baby-crying detector and a screen-capture feature that reads whatever you play over HDMI. We talk about the full spectrum from normal telemetry to creepy to purely malicious, and what the world looks like when anyone can ask these questions of software.
🍳 "Like cooking in somebody else's kitchen": Why everyone's AI harness looks completely different now, and why that's the end of monolith open source rather than open source itself. Also what it feels like to realize you may have already typed your last console.log.
🎮 A shareware game from a "1,000 games" disc: Gabe brought a game from his '90s childhood back to life and added realistic physics, and we get into AI-assisted reverse engineering of games like Super Smash Bros. Melee and gamers who now have more fun building games than playing them.
🪪 "I think we actually might disagree on this one": Whether access to the most capable cyber models should require vetting, given billions of old, unmanaged devices that AI can now reverse engineer. My worry is Thanos gauntlets handed out at 7-Eleven, and Gabe's is a system of haves and have-nots where defenders have to be chosen to get better security while attackers buy access somewhere else.
🍞 "The framing of AI is money and power": Why that framing filters out a lot of people who'd have so much to offer, and how on-ramps that start with curiosity ("you see that toaster over there?") could reach people without leading with AI.
👥 People mentioned:
👤 Gabriel Bernadett-Shapiro, my friend and one of the sharpest minds I know on AI and threat intelligence: https://ul.live/ytgbsgabe
👤 Jason Clinton of Anthropic, whose "crazy 18 months" warning came up: https://ul.live/ytgbsclinton
📚 More on this topic:
🎥 A Conversation with Gabe Bernadett-Shapiro on AI: https://ul.live/ytgbs2023
📖 My post on the reverse engineering angle of AI: https://ul.live/ytgbsrepost
📖 AI is a Gift to Transparency: https://ul.live/ytgbstransparency
📖 IoT Functionality and Personal Privacy are Inversely Correlated: https://ul.live/ytgbsiotprivacy
📖 AI-enabled Self-software: https://ul.live/ytgbsselfsoftware
📖 AI SaaS Replacement is the Fire of Fires: https://ul.live/ytgbsfireoffires
📖 How to Rate the AI We're All Chasing: https://ul.live/ytgbsrateai
📖 Custom Content for Everyone: https://ul.live/ytgbscustomcontent
📖 To Survive AI, We Must Become Creators: https://ul.live/ytgbscreators
📖 Attacker vs. Defender AI Advantage: https://ul.live/ytgbsattackdefend
📖 How Easy It Would Be to Hack You: https://ul.live/ytgbshackyou
🎥 Why We Need Some AI Model Controls: https://ul.live/ytgbsmodelcontrols
🎥 Should Open-Weight AI Be Regulated? A Conversation with Robert Graham: https://ul.live/ytgbsgraham
🌐 Should We Control Open Source AI?: https://ul.live/ytgbscontrolopensource
🔗 Stay connected
📬 Subscribe to the newsletter at:
https://ul.live/ytnl
💬 Join the UL community at:
https://ul.live/ytcommunity
📣 Follow on X:
https://ul.live/ytx
💼 Follow on LinkedIn:
https://ul.live/ytli
📝 I had a blast catching up with my friend Gabriel Bernadett-Shapiro, a Distinguished AI Research Scientist at SentinelOne, after we ran into each other at Unprompted. We geeked out over how reverse engineering went from the most expensive skill in security to totally cheap in about a year, including how Gabe got root on his own TV in a couple of hours and found a hard-coded detector for a baby crying. We also ended up genuinely disagreeing about who should get access to the most capable cyber models, which was a lot of fun.
🧭 What we talk about:
🔓 "Reverse engineering used to be the most expensive skill in security": How AI agents made it cheap almost overnight, like the researcher who used Claude Code agents to turn Call of Duty: Modern Warfare II into a playable build in about three months. Plus what happens to software businesses when you can't safely ship a binary, and Gabe's prediction that we'll end up spending tokens instead of license fees.
📺 "I got root in my TV": Gabe had a model statically analyze his TV's firmware update, found a way in within about two hours, and turned up a baby-crying detector and a screen-capture feature that reads whatever you play over HDMI. We talk about the full spectrum from normal telemetry to creepy to purely malicious, and what the world looks like when anyone can ask these questions of software.
🍳 "Like cooking in somebody else's kitchen": Why everyone's AI harness looks completely different now, and why that's the end of monolith open source rather than open source itself. Also what it feels like to realize you may have already typed your last console.log.
🎮 A shareware game from a "1,000 games" disc: Gabe brought a game from his '90s childhood back to life and added realistic physics, and we get into AI-assisted reverse engineering of games like Super Smash Bros. Melee and gamers who now have more fun building games than playing them.
🪪 "I think we actually might disagree on this one": Whether access to the most capable cyber models should require vetting, given billions of old, unmanaged devices that AI can now reverse engineer. My worry is Thanos gauntlets handed out at 7-Eleven, and Gabe's is a system of haves and have-nots where defenders have to be chosen to get better security while attackers buy access somewhere else.
🍞 "The framing of AI is money and power": Why that framing filters out a lot of people who'd have so much to offer, and how on-ramps that start with curiosity ("you see that toaster over there?") could reach people without leading with AI.
👥 People mentioned:
👤 Gabriel Bernadett-Shapiro, my friend and one of the sharpest minds I know on AI and threat intelligence: https://ul.live/ytgbsgabe
👤 Jason Clinton of Anthropic, whose "crazy 18 months" warning came up: https://ul.live/ytgbsclinton
🎥 A Conversation with Gabe Bernadett-Shapiro on AI: https://ul.live/ytgbs2023
📖 My post on the reverse engineering angle of AI: https://ul.live/ytgbsrepost
📖 AI is a Gift to Transparency: https://ul.live/ytgbstransparency
📖 IoT Functionality and Personal Privacy are Inversely Correlated: https://ul.live/ytgbsiotprivacy
📖 AI-enabled Self-software: https://ul.live/ytgbsselfsoftware
📖 AI SaaS Replacement is the Fire of Fires: https://ul.live/ytgbsfireoffires
📖 How to Rate the AI We're All Chasing: https://ul.live/ytgbsrateai
📖 Custom Content for Everyone: https://ul.live/ytgbscustomcontent
📖 To Survive AI, We Must Become Creators: https://ul.live/ytgbscreators
📖 Attacker vs. Defender AI Advantage: https://ul.live/ytgbsattackdefend
📖 How Easy It Would Be to Hack You: https://ul.live/ytgbshackyou
🎥 Why We Need Some AI Model Controls: https://ul.live/ytgbsmodelcontrols
🎥 Should Open-Weight AI Be Regulated? A Conversation with Robert Graham: https://ul.live/ytgbsgraham
🌐 Should We Control Open Source AI?: https://ul.live/ytgbscontrolopensource
🔗 Stay connected
📬 Subscribe to the newsletter at:
https://ul.live/ytnl
💬 Join the UL community at:
https://ul.live/ytcommunity
📣 Follow on X:
https://ul.live/ytx
💼 Follow on LinkedIn:
https://ul.live/ytli










